Symbolic Reasoning about Large Language Models

This talk by Guy van den Broeck from UCLA explores the integration of symbolic reasoning algorithms with large language models. Discover how traditional AI reasoning approaches using logical and probabilistic deduction can enhance modern language models. Learn about the contrast between current expectations of AI reasoning (generating token sequences resembling chains of thought) and historical symbolic algorithms that derived definite answers from knowledge. The presentation demonstrates how incorporating symbolic reasoning directly into language model architecture enables state-of-the-art capabilities in controllable text generation and alignment. Part of the Theoretical Aspects of Trustworthy AI series at the Simons Institute.